[jira] [Comment Edited] (MADLIB-1270) Unexepcted behavior in vec2cols function

["Frank McQuillan (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/MADLIB-1270?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16575589#comment-16575589 ] 

Frank McQuillan edited comment on MADLIB-1270 at 8/10/18 3:11 PM:
------------------------------------------------------------------

Here is one possible approach:

* Use the max array length to determine the number of columns
* It means some of the generated rows may have NULLs if the array for a particular row is smaller than the max.
* If the optional column_names parameter is used and the number of elements does not match, then throw an error.





was (Author: fmcquillan):
Here is one possible approach:

{code}
IF
feature_names is specified
THEN
make the number of columns equal to the size of the feature_names array

ELSE IF
feature_names is not specified
THEN
make the number of columns equal to the max size of the array in vector_col
{code}

In all cases above, some of the generated rows may have NULLs if the array for a particular row is smaller.




> Unexepcted behavior in vec2cols function
> ----------------------------------------
>
>                 Key: MADLIB-1270
>                 URL: https://issues.apache.org/jira/browse/MADLIB-1270
>             Project: Apache MADlib
>          Issue Type: Bug
>          Components: Module: Utilities
>            Reporter: Rashmi Raghu
>            Priority: Minor
>             Fix For: v1.15.1
>
>
> There is some unexpected behavior when vector column to be split contains different numbers of elements in the vectors. E.g.
> Input table:
> select * from test order by id;
>  id | t 
>  ----+---------
>  1 | \{a,b}
>  2 | \{c,d}
>  3 | \{e,f}
>  4 | \{g,h,i}
>  5 | \{j}
>  (5 rows)
>  
> select madlib.vec2cols('test','test_out_5','t',array['c1','c2','c3'],'id');
>  ERROR: plpy.Error: vec2cols: Mismatch between size of vector_col and number of cols in feature_names.
>  CONTEXT: Traceback (most recent call last):
>  PL/Python function "vec2cols", line 23, in <module>
>  return vec2cols_obj.vec2cols(**globals())
>  PL/Python function "vec2cols", line 149, in vec2cols
>  PL/Python function "vec2cols", line 112, in get_names_for_split_output_cols
>  PL/Python function "vec2cols", line 77, in _assert
>  PL/Python function "vec2cols"
>  
> select madlib.vec2cols('test','test_out_5','t',array['c1','c2'],'id');
>  vec2cols 
>  ----------
> (1 row)
> select * from test_out_5 order by id;
>  id | c1 | c2 
>  ----++--------
>  1 | a | b
>  2 | c | d
>  3 | e | f
>  4 | g | h
>  5 | j | 
>  (5 rows)
>  
>  
> select madlib.vec2cols('test','test_out_6','t',array['c1'],'id');
> ERROR: plpy.Error: vec2cols: Mismatch between size of vector_col and number of cols in feature_names.
>  CONTEXT: Traceback (most recent call last):
>  PL/Python function "vec2cols", line 23, in <module>
>  return vec2cols_obj.vec2cols(**globals())
>  PL/Python function "vec2cols", line 149, in vec2cols
>  PL/Python function "vec2cols", line 112, in get_names_for_split_output_cols
>  PL/Python function "vec2cols", line 77, in _assert
>  PL/Python function "vec2cols"
>  
> --- Update-----
> There are a couple of decisions to be made regarding supporting arrays of different lengths:
> -If we choose the array with maximal length in the vector_col, what do we do if the user's passed-in feature_names does not have the same number of elements?
> -What are the performance issues with looking through our vector_col for the array with maximal length?
> -How will we handle default feature names: will we create a feature name for every element of the longest array entry?



--
This message was sent by Atlassian JIRA
(v7.6.3#76005)